1: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

2: %

3: %Dear Editor, 

4: % 

5: %please find enclosed the manuscript "Quantum phase transition 

6: %in the Frenkel-Kontorova chain: from pinned instanton glass to 

7: %sliding phonon gas" which we submit for publication in Phys. Rev. E

8: %as a regular article. 

9: %The paper contains 18 figures attached in ps format. 

10: %

11: %All correspondence should be addressed to:

12: %Dima Shepelyansky

13: %Laboratoire de Physique Quantique

14: %UMR 5626 du CNRS

15: %Universite Paul Sabatier

16: %31062 Toulouse Cedex 4 

17: %France

18: %

19: %e-mail: dima@irsamc.ups-tlse.fr

20: %fax: +33-5-61556065

21: %tel: +33-5-61556068

22: %http://w3-phystheo.ups-tlse.fr/~dima

23: %

24: %Sincerely yours, 

25: %

26: %Oleg Zhirov, Giulio Casati and  Dima Shepelyansky 

27: %

28: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

29: \documentclass[twocolumn,english,aps,pre,showpacs]{revtex4}

30: \usepackage[T1]{fontenc}

31: \usepackage[latin1]{inputenc}

32: \usepackage{float}

33: \usepackage{amsmath}

34: \usepackage{graphicx}

35: \usepackage{amssymb}

36: %\usepackage[active]{srcltx} 

37: 

38: \makeatletter

39: 

40: \newcommand{\Ci}[1]{ }

41: \newcommand{\Cf}[1]{ }

42: %\newcommand{\Ci}[1]{\textbf{C$_i$(#1)}}

43: %\newcommand{\Cf}[1]{\textbf{C$_f$(#1)}}

44: 

45: \usepackage{babel}

46: \makeatother

47: \begin{document}

48: 

49: \preprint{preprint}

50: 

51: 

52: \title{Quantum phase transition in the Frenkel-Kontorova chain:\\

53: from pinned instanton glass to sliding phonon gas}

54: 

55: 

56: \author{O.V. Zhirov}

57: \email{zhirov@inp.nsk.su}

58: \affiliation{Budker Institute of Nuclear Physics, 630090 Novosibirsk, Russia}

59: 

60: \author{G. Casati}

61: \email{giulio.casati@uninsubria.it}

62: \affiliation{International Center for the Study of Dynamical 

63: Systems, Universit\`a degli Studi dell'Insubria and\\

64: Istituto Nazionale per la Fisica della Materia, 

65: Unit\`a di Como, Via Valleggio 11, 22100 Como, Italy and\\

66: Istituto Nazionale di Fisica Nucleare, 

67: Sezione di Milano, Via Celoria 16, 20133 Milano, Italy}

68: 

69: \author{D.L. Shepelyansky}

70: \email{dima@irsamc.ups-tlse.fr}

71: \affiliation{Laboratoire de Physique Quantique, UMR 5626 du CNRS, Universit\'{e}

72: Paul Sabatier, 31062 Toulouse, France}

73: 

74: 

75: \date{October 17, 2002}

76: 

77: \begin{abstract}

78: We study analytically and numerically the one-dimensional quantum Frenkel-Kontorova

79: chain in the regime when the classical model is located in the pinned 

80: phase characterized by the gaped phonon excitations and devil's staircase.

81: By extensive quantum Monte Carlo simulations we show 

82: that for the effective  Planck constant $\hbar$  

83: smaller than the critical value

84: $\hbar_c$ the quantum chain is in the pinned instanton glass phase.

85: In this phase

86: the elementary excitations have two branches:

87: \emph{phonons}, separated from zero energy by a finite gap, and \emph{instantons} which

88: have an exponentially small excitation energy. At  $\hbar=\hbar_c$

89: the quantum phase transition

90: takes place and for $\hbar>\hbar_c$ the pinned instanton glass is transformed into the

91: sliding phonon gas with gapless phonon excitations.

92: This transition is accompanied by the divergence of the spatial correlation

93: length and appearence of sliding modes at $\hbar>\hbar_c$.

94: \end{abstract}

95: \pacs{05.45.-a, 63.70.+h, 61.44.Fw }

96: \maketitle

97: 

98: 

99: 

100: \section{introduction}

101: 

102: The Frenkel-Kontorova (FK) model \cite{FK38} describes a one-dimensional

103: chain of atoms/particles with harmonic couplings placed in a periodic

104: potential. This model was introduced more than sixty years ago with

105: the aim to study crystal dislocations \cite{FK38,Naba67}. It was

106: also successfully applied later to the description of commensurate-incommensurate

107: phase transitions \cite{VPaAT84}, epitaxial monolayers on the crystal

108: surface \cite{SY71}, ionic conductors and glassy materials \cite{Piet81,Au78,Au83a}

109: and, more recently, to charge-density waves \cite{Flor96} and dry

110: friction \cite{Brau97,Cons00}. Despite the fact that the relevant

111: phenomena are at the atomic scale, all these works are based essentially

112: on the classical approach. The first study of quantum effects was done 

113: twelve years ago \cite{Borg89,Borg90,Borg89d} with the attempt

114: to understand the highly nontrivial quantum ground state of the Frenkel-Kontorova model

115: in the regime when the classical ground state is characterized

116: by the fractal ``devil staircase'' \cite{Au78}. These studies were

117: extended in 

118: \cite{Berm94,Berm97} and later the quantum dynamics at different values

119: of quantum parameter $\hbar $ was studied in \cite{Hu98,Hu99,Hu01,Hu01r,Ho01}. 

120: 

121: 

122: The physical properties even of the classical FK model are very rich 

123: and nontrivial. In 1978 Aubry discovered \cite{Au78} a new type of ground

124: state which has fractal properties known as ``devil's staircase''. 

125: In fact, the equilibrium positions of atoms in the FK chain 

126: are described \cite{Au78} by the well known Chirikov standard map \cite{Chir79},

127: which describes generic properties of chaotic Hamiltonian dynamics.

128: The density of particles in the equilibrium

129: state determines the rotation number of the invariant curves of the

130: map, while the amplitude of the periodic potential in the FK model

131: gives the value of the dimensionless parameter $K$. According to

132: the known properties of the map, it follows that there exist two phases

133: of the chain, the ``sliding'' phase and the ``pinned''

134: phase. Indeed at $K<K_{c}$ the Kolmogorov-Arnold-Moser (KAM) curves are smooth and 

135: the chain can easily slide along the potential. This implies the existence of

136: a zero phonon mode. In contrast, at $K>K_{c}$ the KAM curves are

137: destroyed and replaced by an invariant Cantor set which is called

138: cantorus \cite{Au78,Perc79,Au83b,Au83c,Mac84}. In this ``pinned'' phase the chain cannot

139: slide being kept by a finite Peierls-Nabarro barrier and the phonon

140: spectrum is separated from zero by a finite gap. In this paper we consider only

141: the gaped/pinned phase with  $K>K_{c}$.

142: 

143: It is important to stress that in the pinned phase, besides the equilibrium state

144: of minimal energy, there exists a lot of other \emph{equilibrium} configurations,

145: corresponding to \emph{local} minima of the potential with energies

146: very close to the minimal energy, i.e. the energy of the ground state

147: \cite{Sch84,Sch85,Zhir01a}. The total amount of these states (configurational

148: excitations) grows exponentially both with the length $L$ of the

149: chain, as well as with the parameter $K$ \cite{Zhir01a}. Moreover, a great number of

150: them is practically degenerate since their energy separation from

151: the ground state is exponentially small. 

152: 

153: In the classical limit all these configurational excitations

154: are stable, while in the quantum case they become metastable due to tunneling

155: between different exponentially degenerate minima of the potential. As 

156: a result one may expect that the true eigenstates of the Hamiltonian, including

157: the ground state, are built as superpositions of classical configurational

158: states. Such a nontrivial structure of the quantum ground state of the chain has

159: a profound  analogy with the famous vacuum of Quantum

160: Chromodynamics (QCD) \cite{Shur98} in which tunneling transitions between different,

161: practically degenerate, states are known as ``instantons''.

162: This analogy between the two problems is very useful and implies 

163: that many of the methods developed in lattice QCD studies may

164: be applicable to quantum FK model.

165: 

166: Dynamical low energy excitations in the classical FK chain consist

167: only of phonon modes describing small vibrations around a classical minimum

168: of the chain potential energy. In the quasiclassical regime, where  

169: the effective dimensionless Planck constant $\hbar$ is very small,

170: the quantization of the FK chain can be reduced to a quantization of phonon

171: modes only. Indeed the time of tunneling between different minima of

172: the potential energy is exponentially large, compared to periods of

173: the vibrations, and the phonon modes are decoupled from the tunneling

174: modes (instantons). 

175: This reminds the situation in the QCD \cite{Shur98}

176: where the quasiclassical regime for instantons appears at small distances

177: on which instantons are also decoupled from other excitations like 

178: quarks and gluons.

179: 

180: In this regime the tunneling transitions are very slow, instantons are local

181: and frozen in space. Due to exponential degeneracy of chain configurations the

182: model reveals a glass-like structure of instantons randomly distributed   

183: along the chain. We will call this phase the ``instanton glass''.

184: 

185: One may expect that the phase structure changes significantly with the increase 

186: of $\hbar$ when the

187: tunneling time $t\sim \exp (S_{in}/\hbar )$ ($S_{in}$ is the instanton

188: action) becomes comparable to the inverse frequency of phonons. In this

189: case phonon and instanton excitations can strongly influence each other

190: due to anharmonicity of the periodical

191: potential. This may lead to quantum melting of the instanton glass phase

192: and transition to another phase which appears above some critical value 

193: $\hbar>\hbar_c$.

194: 

195: The existence of another quantum phase can be argued in the following way.  

196: At sufficiently large  $\hbar \geq \hbar _{c}$ the kinetic energy of

197: quantum particle $E_{k}\sim \hbar^{2}/2m(\Delta x)^{2}$ starts to exceed

198: the height $U$ of the periodic potential in which the chain is placed

199: (here $\Delta x$ is the mean particle separation in the chain which is comparable

200: with the period of the potential, $m$ is the particle mass). 

201: The condition $E_{k}\sim U$ gives $\hbar_c\sim\Delta x\sqrt{mU}$.

202: As a result, for $\hbar>\hbar_c$ the chain turns from the \emph{pinned} 

203: to the \emph{sliding} phase. In this regime tunneling is replaced by direct 

204: propagation above the barrier.

205: Hence, the instanton-like motion is replaced by phonon-like motion corresponding 

206: to a new phase with gapless phonon excitations. This regime can be considered as 

207: sliding quantum phase similar to classical sliding regime at $K<K_c$ \cite{Au83b,Au83c}.

208: Qualitative different types of the behavior of quantum FK model at small and large

209: $\hbar$ values were already seen in the first numerical studies \cite{Borg89,Borg90}.  

210: In summary, at $\hbar =\hbar _{c}$

211: one may expect a \emph{quantum phase transition}, with qualitative

212: rearrangement of the spectrum of elementary quasiparticle excitations.

213: 

214: 

215: In this work, devoted to a comprehensive numerical study of the transition

216: we have just outlined, we present a complete picture of low energy

217: excitations in the quantum FK chain. We also find that this

218: transition is highly nontrivial and the model reveals a non-analytical

219: behavior which can partially explain the failure of simple analytical

220: approaches developed in \cite{Berm94,Hu98,Hu99,Ho01}. It should be

221: stressed also that our results have relevance to a wider class of

222: quantum systems, like e.g. quantum spin glass or other disordered systems 

223: with interactions. Indeed, the existence of highly degenerate classically 

224: stable configurations is a general property of such systems.

225: 

226: 

227: The paper is organized as follows. In Section \ref{sec:the-fk-model}

228: we outline the model, its quantum features and the main points of our

229: numerical approach. In Section \ref{sec:elementary-excitations.}

230: we study  elementary excitations of the chain at different

231: values of $\hbar $ and show that there exists a structural rearrangement of

232: the excitation spectrum at certain $\hbar = \hbar_c$. 

233: A detailed analysis of this rearrangement, given in Section \ref{sec:QFT}, 

234: indicates that we have a quantum phase transition. Our results are 

235: summarized in Section \ref{sec:Conclusions}.

236: 

237: 

238: \section{The quantum Frenkel-Kontorova model}

239: \label{sec:the-fk-model}

240: 

241: \subsection{Definitions and outline of quantum features}

242: 

243: The model describes a one-dimensional chain of particles with harmonic

244: couplings placed in a periodic potential. The Hamiltonian reads:

245: \begin{equation}

246:   H=\sum _{i=0}^{s}\left[\frac{P_{i}^{2}}{2m}+\alpha\frac{(x_{i}-x_{i-1})^{2}}{2}-

247:   \beta\cos(x_{i}/d)\right]

248: \label{H_FKorig}

249: \end{equation}

250:  where $s$ is the number of particles, $P_{i}$ and $x_{i}$ are

251: their momenta and coordinates, and in the quantum case 

252: $P_{i}=-i\hbar\partial/\partial x_i$. In this paper we use  

253: units $m=d=\alpha=1$, that correspond to dimensionless Hamiltonian 

254: (see for details e.g. \cite{Borg89,Borg90,Berm94}):

255: \begin{equation}

256:   H=\sum _{i=0}^{s}\left[\frac{P_{i}^{2}}{2}+\frac{(x_{i}-x_{i-1})^{2}}{2}-

257:   K\cos(x_{i})\right]

258: \label{H_FK}

259: \end{equation}

260: with a dimensionless parameter $K=\beta/\alpha d^2$ and the

261: dimensionless Planck constant $\hbar$ is measured in units of $d^2\sqrt{m \alpha}$. 

262: In this way, $K$ is the chaos parameter in the Chirikov standard map \cite{Chir79,Au83b}.

263: We use the standard boundary conditions

264: \begin{equation}

265:   x_{0}=0,\quad x_{s}=L\, ,\label{Bcond}

266: \end{equation}

267: where the chain length $L=2\pi \cdot r$ consists of $r$ periods/wells 

268: of the external field. We analyze the standard case of golden mean ratio corresponding 

269: to $r/s\to(\sqrt{5}-1)/2$ (see, \cite{Borg89,Borg90,Zhir01a}).

270: The potential energy of the chain

271: \begin{equation}

272:   U(\{x\})=\sum _{i=0}^{s}\left[\frac{(x_{i}-x_{i-1})^{2}}{2}-

273:   K\cos (x_{i})\right]

274: \label{Upot}

275: \end{equation}

276: has a large number of minima, corresponding to different possible

277: distributions of particles among the wells. The classical 

278: ground state (absolute minimum of the potential energy) is

279: characterized by some special ordering discovered by Aubry 

280: \cite{Au78,Au83b,Au83c}. In addition, there are also local

281: minima, known as ``configurational excitation'' states\cite{Be80}.

282: Many of them have exponentially small energy separations from

283: the energy of the classical ground state \cite{Zhir01a}. In the classical

284: case all these states are well defined (distinguishable) and absolutely

285: stable.

286: 

287: In contrast, in the \emph{quantum} world these states are \emph{metastable}

288: due to quantum tunneling. Since they are practically degenerate with

289: the classical ground state the actual \emph{quantum} ground state

290: is built by \emph{many} of them. The admixture of metastable classical 

291: configurations in the quantum ground state was first discussed in \cite{Berm94,Berm97}. 

292: Let us summarize below 

293: the most important aspects of the quantum ground state, related to

294: the tunneling between these configurations.

295: 

296: In the quasiclassical region, the transition amplitudes between different

297: metastable configurations are exponentially small. Hence,

298: the ground state wave function is (with exponential accuracy) a sum

299: of \emph{non}-overlapping parts, each referring to a particular classical

300: configuration. Any average over the quantum ground state is (within

301: the same accuracy) a weighted sum of averages over relevant classical

302: configurations. Note that in this limit the main contribution to quantum

303: motion comes from phonons which characterize small vibrations around a 

304: classical equilibrium 

305: configuration. In fact, the phonon spectrum is only weakly sensitive

306: to the choice of specific configuration \cite{PhonFK} and 

307: the influence of tunneling processes on the global (thermodynamical) 

308: properties of the chain is expected to be small. In particular,

309: the phonon spectrum is still characterized by a phonon gap similar to 

310: the classical case. In this regime the tunneling 

311: transitions between metastable configurations are very slow comparing to 

312: phonon frequencies and can be considered as well separated instantons.

313: 

314: At higher values of $\hbar $ the tunneling rate increases and becomes

315: comparable with the frequency of phonons. In this regime phonon

316: oscillations are large and essentially anharmonic and therefore the

317: known analytical approaches \cite{Berm94,Hu98,Hu01r,Ho01} based on

318: the gauss-like wave function profile are not quite adequate. Instead, 

319: interactions between instantons and phonons come into play here.

320: As it will be shown later,  this leads to a new sliding phase appearing

321: at $\hbar>\hbar_c$. 

322: 

323: In fact, the first effects of quantum tunneling between classical configurations

324: were already seen in the numerical studies \cite{Borg90} (see Fig.4b). However, 

325: they were not attentively analyzed and the mixture of classical metastable 

326: configurations induced by quantum tunneling was not discussed. 

327: Certainly, tunneling processes affect significantly the average positions of 

328: particles at large time scales. Therefore the quantities based on mean expectation 

329: values of particle positions are not adequate even in the deep quasiclassical regime. 

330: Hence the basic concepts of classical treatment, like hull \cite{Au78,Au83b}

331: and g-functions \cite{Borg89,Borg90}, are also not adequate in this quantum regime.

332: The point is that in the case of particle tunneling between

333: two classical equilibrium positions the expectation value

334: does not correspond to any \emph{probable} (in the classical sense)

335: particle position. This is evident from the analogy with the two-well

336: potential problem, where the mean expectation value of a particle

337: position coincides with the \emph{top} of the barrier. 

338: 

339: 

340: \subsection{Path integral and numerical simulations}

341: \label{sub:Path-Sim}

342: 

343: Tunneling effects are best understood in the Feynman path integral

344: formulation of quantum mechanics\cite{Poly77}. In particular, the

345: transformation to ``Euclidean'' time variable $\tau$: $t=i\tau$

346: makes the tunneling transition, or \emph{instanton,} local in time

347: $\tau$. In the quasiclassical regime the tunneling probability

348: is very small and therefore the mean separation between instantons 

349: is large compared to their size. This allows to use the approximation

350: of dilute instanton gas \cite{Poly77}. 

351: 

352: Feynman path formulation in the ``Eucleudean'' time 

353: allows direct numerical simulations of quantum systems with

354: probabilistic treatment \cite{Creu81} of the path integral:

355: \begin{equation}

356:   Z=\int Dx[\tau ]\exp (-\frac{1}{\hbar }S[x(\tau )]),

357: \label{FeyP}

358: \end{equation}

359: where the ``Eucledian'' action

360: \begin{equation}

361:   S[x(\tau )]=\int _{0}^{\tau _{0}}d\tau \sum_i

362:   \left(\frac{\dot{x}_{i}^{2}}{2}+\frac{(x_{i}-x_{i-1})^{2}}{2}-

363:   K\cos (x_{i})\right)

364: \label{Sc}

365: \end{equation}

366: has the same form as a total energy of a chain of particles in the usual 

367: real time variable, integrated over some time interval $\left[0,\tau _{0}\right]$.

368: This is equivalent to the consideration of the quantum system at some finite

369: temperature\cite{Creu81}

370: \begin{equation}

371:   T=\hbar /\tau _{0}  %\label%{eq:Tempr}

372: \label{eq:Temperature} 

373: \end{equation}

374: 

375: We assume periodic boundary conditions in the \emph{time}

376: direction, which correspond to a path closed on a torus 

377: \begin{equation}

378:   x_{i}(\tau _{0})=x_{i}(0),\quad i=1,\ldots ,s-1,

379: \label{TbndCnd}

380: \end{equation}

381: and integrate over the initial conditions $x_{i}(0)$ in order to

382: restore homogeneity of paths along the time torus. This allows to

383: improve the statistics in data measurements by averaging the data

384: along the torus.

385: 

386: The numerical simulation of the path integral (\ref{FeyP}) needs

387: discretization of the time variable $\tau _{n}=\Delta \tau \cdot n$,

388: $n=1,\ldots ,N$ by splitting the time interval $\tau_ {0}$ into $N$

389: steps of size $\Delta \tau =\tau_0 /N$. As a result the original 

390: time-continuous model (\ref{H_FK}) turns into a 2-dimensional lattice 

391: model with the action

392: \begin{eqnarray}

393:   S & = & \sum _{n=0}^{N}\sum _{i=1}^{s}\left[\right.

394:         \frac{(x_{i,n+1}-x_{i,n})^{2}}{2\Delta \tau }+\nonumber \\

395:     &   & +\Delta \tau \frac{(x_{i+1,n}-x_{i,n})^{2}}{2}-

396:         \Delta \tau \cdot K\cos x_{i,n}\left.\right].

397: \label{eq:S_Lat}

398: \end{eqnarray}

399: 

400: The time step $\Delta \tau $ should be chosen small enough to approach

401: the continuous limit of the original model (\ref{H_FK}). This leads

402: to the following requirements. At any time step $\Delta \tau $

403: the path variable $x_{i}$ jumps by a random shift 

404: $\Delta x_{i}\sim \sqrt{\hbar \Delta \tau }$. The first obvious requirement 

405: is that the shift $\Delta x_{i}$ should be small compared to the spatial scale 

406: of the potential energy variation.

407: Another requirement comes from the standard derivation of the path

408: integral \cite{Creu81,Shur84}: the potential energy terms should

409: be small compared to the kinetic energy:

410: \begin{equation}

411:   \Delta \tau \cdot \left(\frac{(x_{i+1,n}-x_{i,n})^{2}}{2}-K\cos x_{i,n}\right)

412:   \ll \frac{(x_{i,n+1}-x_{i,n})^{2}}{2\Delta \tau }.

413: \label{P<<K}

414: \end{equation}

415: The latter condition makes our lattice \emph{anisotropic} with respect

416: to spatial/time directions. In spite of the common belief \cite{cQFT97}

417: on the equivalence between quantum 1-dimensional chains and classical

418: 2-dimensional statistical (e.g. \textsl{XY}) models, the above discussion

419: shows that some particular care has to be taken.

420: 

421: An accurate treatment of particle tunneling between two wells in the chain

422: requires some modifications of the action (\ref{eq:S_Lat}).

423: For a finite (not very small!) time step $\Delta \tau $, there exists

424: a probability for a particle to jump over the potential barrier in

425: \emph{one} time step\cite{Shur84}. The corresponding path contribution

426: to the action will be strongly underestimated by eq.(\ref{eq:S_Lat})

427: and, respectively, the probability of tunneling will be too high.

428: To cure this problem we use in our simulation an improved \cite{Shur84}

429: version of action (\ref{eq:S_Lat}), with the potential energy term

430: $\Delta \tau \cdot U(x_{n})$ replaced by the integral 

431: $\Delta \tau \cdot \int _{x_{n}}^{x_{n+1}}U(x)dx/(x_{n+1}-x_{n})$

432: along a straight line that links subsequent (in time) points $(x_{n},x_{n+1})$

433: of the particle path. This significantly improves the accuracy of 

434: numerical simulations.

435: 

436: For simulations of path ensembles we use the standard Metropolis algorithm

437: \cite{Metr53}. Each iteration looks as follows: at any fixed time

438: slice at number $n$ we update sequentially the particles coordinates

439: $x_{i,n}$, $i=1,\ldots ,s$; then we go to the next time slice: $n\rightarrow n+1,$

440: and so on. 

441: 

442: The system has two principal characteristic relaxation time scales

443: that are originated by different underlying processes. Basically one

444: can estimate the number of iterations as $N_{it}\sim 1/\omega ^{(min)}\Delta \tau $,

445: where $\omega^{(min)}$ is the lowest frequency relevant

446: to the process and $\Delta \tau $ is time discretization step. The

447: shortest scale is related to the path relaxation with respect

448: to main phonon modes, and is typically of the order of a few tens of iterations

449: in the quasiclassical regime, where phonons have a gap of

450: order unity. Another time scale is much larger and is determined by the smallest

451: frequency related to either the tunneling rate between

452: different classical configurations, or to the lowest phonon

453: frequency available in the system for $\hbar \geq \hbar _{c}$.

454: 

455: For the initial state we choose the Aubry classical ground state. Then, applying 

456: iterations, we generate a path ensemble. To be sure that the system does actually 

457: relax to statistical equilibrium with respect to the slowest processes described above, 

458: we control the mean number of particle path crosses a top of the potential 

459: barrier and we discard all configurations in the ensemble until this quantity

460: stabilizes. All computations are done for the chaos parameter $K=5$.

461: The required number of iterations to reach relaxation is very sensitive 

462: to the value of quantum parameter $\hbar $: for example at $\hbar =3$ this number is

463: $N_{it}\sim (2\div 4)10^{2}$ , while  at $\hbar=1$ it is  of the order of $10^{5\div 6}$.

464: This explains why in the first studies \cite{Borg89,Borg90} done at

465: $N_{it}\leq10^4$ many details at $\hbar \leq 2$ were not seen.

466: We study chains with up to 233 particles.

467: 

468: \section{elementary excitations}

469: \label{sec:elementary-excitations.}

470: 

471: The most important information about the quantum system is contained

472: in its spectrum of low-lying elementary excitations. Being the net

473: manifestation of system internal structure it reflects any structural

474: transition which can occur in the system, and it provides a complete

475: description of low-temperature thermodynamic and kinetic

476: properties. We extract this spectrum using a novel approach based

477: on the analysis of Fourier spectrum of Feynman paths. This approach

478: provides a most direct way to see and resolve different excitations

479: in the system. Then  we compare our method with the more traditional

480: one, based on the study of time correlation functions. This comparison 

481: provides a self-consistency check and demonstrates the advantages 

482: of our method.

483: 

484: 

485: \subsection{Spectral properties of Feynman paths}

486: \label{sub:SpFunc}

487: 

488: In classical nonlinear dynamics, the Fourier analysis of trajectories plays

489: a key role in understanding of periodic motion of complex systems. In

490: a similar way,

491: the spectral characteristics of Feynman paths are closely related

492: to the properties of elementary excitations in quantum systems. We start

493: our studies from the quasiclassical limit $\hbar \to 0$, where

494: this relation is exact, and extend them to higher values of

495: $\hbar $.

496: 

497: Let us consider the Fourier image of the path variable $x_{i}(\tau )$:

498: \begin{equation}

499:   a_{i}(\omega _{m}) = \frac{1}{\sqrt{\tau _{0}}}\int _{0}^{\tau _{0}}d\tau 

500:   \, x_{i}(\tau )\, \exp (i\omega _{m}\tau )

501: \label{eq:four_x}

502: \end{equation}

503: where $\omega _{m}=m\varpi $, $\varpi \equiv 2\pi /\tau _{0}$, and

504: $-\infty <m<\infty $. The path variable $x_{i}(\tau )$ is real,

505: therefore $a_{i}(-\omega _{m})=(a_{i}(\omega _{m}))^{\ast }$. To

506: get insight into the physical content of this quantity let us consider

507: the quasiclassical regime $\hbar \ll 1$. Then, for small variations

508: $x_{i}(\tau )=\overline{x}_{i}+\delta x_{i}(\tau )$ around the classical

509: static trajectory $\{\overline{x}_{i}\}$, one can expand the action

510: (\ref{Sc}) up to second order terms in $\delta x_{i}(\tau )$. Next,

511: using the spectral expansion for $\delta x_{i}(\tau )$ and performing

512: the integration over $\tau $, one gets:

513: \begin{eqnarray}

514:  S & = & S_{0}[\overline{x}]+\nonumber \\

515:    & + & \sum _{m}\sum _{i,k}\frac{1}{2}(\omega _{m}^{2}\delta _{ik}-

516:    \Omega _{ik}^{2})a_{i}(-\omega _{m})a_{k}(\omega _{m})\label{eq:Sexpa}\\

517:    \Omega _{ik}^{2} & = & \delta _{ik}\left(2+K\cos (x_{i})\right)-

518:    \delta _{i,k-1}-\delta _{i-1,k}\label{eq:Om2}

519: \end{eqnarray}

520: Now, by the transformation to normal modes 

521: $A^{(l)}(\omega _{m})=\sum _{i}V_{i}^{(l)}a_{i}(\omega _{m})$,

522: where eigenvectors $V_{i}^{(l)}$ satisfy the equation 

523: $\Omega _{ik}^{2}V_{k}^{(l)}=\nu _{l}^{2}V_{i}^{(l)}$

524: one gets the standard representation of the action as a sum of independent

525: phonon modes:

526: \begin{equation}\label{eq:Om3}

527: S=S_{0}[\overline{x}]+\sum _{m}\sum _{l}\frac{1}{2}(\omega _{m}^{2}+\nu _{l}^{2})\left|A^{(l)}(\omega _{m})\right|^{2}

528: \end{equation}

529: Finally, the path integral (\ref{FeyP}) turns into a product of ordinary

530: integrals

531: \begin{eqnarray*}

532: Z & = & \int \prod _{l=1}^{s-1}dA^{(l)}(0)\exp \left(-\nu _{l}^{2}\left|A^{(l)}(0)\right|^{2}/2\hbar \right)\times \\

533:  &  & \quad \prod _{m=1}^{\infty }d{\textrm{Re}}A^{(l)}(\omega _{m})d{\textrm{Im}}A^{(l)}(\omega _{m})\times \\

534:  &  & \quad \exp \left(-(\omega _{m}^{2}+\nu _{l}^{2})\left|A^{(l)}(\omega _{m})\right|^{2}/\hbar \right),

535: \end{eqnarray*}

536: and one arrives to the well known result for the correlator of free phonon

537: modes: 

538: \begin{equation}

539:  \left\langle A^{(l)}(\omega _{m})A^{(l)\ast }(\omega _{m'})\right\rangle =

540:  \frac{\hbar \delta _{mm'}}{(\omega _{m}^{2}+\nu _{l}^{2})}.

541: \label{eq:corrAA}

542: \end{equation}

543: We note that this result is obtained from Eq.(\ref{Sc}) where the

544: action is \emph{continuous} in time variable. In the discretized version

545: (\ref{eq:S_Lat}) the number of harmonics is finite: $\left|m\right|=0,1,\ldots ,M$,

546: where $M=\tau _{0}/2\Delta \tau $. It can be shown that the only modification 

547: induced by discretization is the replacement in (\ref{eq:corrAA}):

548: $\omega _{m}\rightarrow \widetilde{\omega }_{m}=(2\omega _{M}/\pi )\sin (\pi \omega _{m}/2\omega _{M})$,

549: $\omega _{M}\equiv M\varpi $. As a result, the spectral function 

550: for phonons is given by

551: \begin{equation}

552:  F^{(l)}(\widetilde{\omega }_{m})\equiv 

553:    \left\langle \left|A^{(l)}(\omega _{m})\right|^{2}\right\rangle =

554:    \frac{\hbar }{(\widetilde{\omega }_{m}^{2}+\nu _{l}^{2})}.

555: \label{eq:Fw}

556: \end{equation}

557: Hereafter, instead of $\omega_m$ we assume its discretized version 

558: $\widetilde{\omega}_m$, and in the following  the tilde will be omitted.

559: 

560: The expression (\ref{eq:corrAA}) is the well known Wick rotated 

561: Green function (in the frequency representation) for a single free particle 

562: in the phonon field theory, which has in our case one spatial dimension.

563: 

564: In the quasiclassical regime the amplitudes of phonon oscillations are small and 

565: the interactions between phonons due to anharmonicity of the Hamiltonian (\ref{H_FK}) 

566: are negligible. At higher $\hbar$, the amplitudes of phonon vibrations grow 

567: as $\hbar^{1/2}$ and their interactions become more important. In general, 

568: interactions can essentially modify the Green function (\ref{eq:corrAA}) for 

569: phonon excitations.  This actually happens for $\hbar>\hbar_c$ where the spectrum

570: of excitations is significantly changed.

571: However, one may expect that the  spectral function of elementary excitation 

572: remains of the same form 

573: \begin{equation}

574:   F^{(l)}(\omega _{m})=

575:   f\frac{\hbar }{(\omega_{m}^{2}+\nu _{l}^{2})},

576: \label{eq:Fw_renorm}

577: \end{equation}

578: which differs from (\ref{eq:corrAA}) by the renormalized frequency value $\nu _{l}$

579: and by an overall renormalization factor $f$, in analogy with the Green function 

580: behavior in renormalizable quantum field theories (see, e.g. \cite{Shur98}). 

581: In fact, this idea is well supported by numerical data.

582: 

583: An extended elementary excitation involves all particles in the chain. In turn,

584: the Fourier harmonics of any particle coordinate in the chain are a

585: sum of contributions of many elementary excitations

586: \begin{equation}

587:  \left\langle \left|a_{i}(\omega _{m})\right|^{2}\right\rangle =

588:  \sum _{j}f^{(j)}_i\frac{\hbar }{(\omega_{m}^{2}+\nu _{j}^{2})},

589: \label{eq:Fw_x_i}

590: \end{equation}

591: where the sum goes over all chain excitations.

592: In the deep quasiclassical case the main contribution comes from phonons modes, 

593: and the sum goes over phonon modes $l$, with 

594: $f^{j}_i\to\left|V_{i}^{(l)}\right|^{2}$ and $\nu_{j}\to\nu_l$.

595: 

596: The goal of our study here is a complete picture of low-lying quantum

597: excitations in the chain, both for low and high values of $\hbar $.

598: In fact, the spectrum of low-lying excitations is crucially dependent

599: on $\hbar $: there are domains with a \emph{qualitatively} different

600: behavior. In order to illustrate this, let us consider the amplitude

601: of quantum motion of particles in the chain, given by Eq.(\ref{eq:Fw_x_i}).

602: It is seen that the contributions of low-frequency modes(small $\nu_l$) dominate in the

603: limit $\omega _{m}\rightarrow 0$, provided that their wave-function

604: profiles $V_{i}^{(l)}$ are not small at the particle position $i$.

605: Therefore the spectral function (\ref{eq:Fw_x_i}), computed in this limit,

606: gives a rough estimate for the frequency $\nu_l$ of the lowest mode. 

607: 

608: %

609: \begin{figure}[th]

610: \includegraphics[  clip,

611:   width=105mm,

612:   height=85mm,

613:   keepaspectratio,

614:   angle=270,

615:   origin=c]{fig1.ps}

616: \vskip -10mm

617: \caption{\label{fig:overview}Dependence of the amplitude of the lowest Fourier

618: harmonic $a_{i}\equiv a_{i}(\omega_1)$  on the particle

619: position $i$ in the chain at different $\hbar $. 

620: Here $\omega_1=\varpi =2\pi /\tau _{0}$, $m=1$. The chain parameters

621: are $s/r=89/55$, $K=5$, $\tau _{0}=80$. Typical number of iterations 

622: is $(1.5\div 5)\cdot 10^5$ at each

623: value of $\hbar$.}

624: \end{figure}

625: 

626: 

627: In Fig.\ref{fig:overview} the amplitude of the \emph{lowest} Fourier

628: harmonic $a_{i}(\omega_1)$ with $\omega_1=\varpi =2\pi /\tau _{0}$ is plotted

629: as a function of the particle position $i$ at different $\hbar =0.6-8$.

630: One can see that the whole interval of $\hbar $ splits naturally in

631: \emph{three} regions of qualitatively different behavior: (i) the

632: quasiclassical region $\hbar \lesssim 1$ where the amplitudes $a_i(\omega_1)$ of

633: the harmonics are very small and depend on the particle positions in some regular way;

634: (ii) the transition region $1\lesssim \hbar \lesssim 2$,

635: where this dependence is highly irregular and interactions between instantons and

636: phonons are important; and (iii) the region $\hbar \gtrsim 2$,

637: where this dependence becomes regular again and where, as we shall see below, 

638: a new phonon branch appears. 

639: Let us note, that at $\hbar<1$ the regular structure along the chain

640: is quasiperiodical, which reflect a fact that a classical chain is built of ``bricks'' 

641: of two principal sizes \cite{Zhir01a}. Above $\hbar\approx 2$ bricks are ``melted''

642: and chain properties become even more homogeneous along the chain. In the 

643: intermediate region irregular peaks come from different non-overlapping

644: instantons contributions, which as 

645: any tunneling effects are highly sensitive to small variations of potential barriers.

646: Below their contributions are exponentially small, and above, as we see further, they 

647: overlap and form new \textit{sliding} phase of the system.

648: 

649: 

650: In the following we analyze in detail these regions, corresponding to different 

651: intervals of $\hbar$.

652: 

653: 

654: \subsection{Quasiclassical region $\hbar \lesssim 1$}

655: \label{sub:Quasiclassical}

656: 

657: For $\hbar\lesssim 1$ the tunneling between different 

658: metastable classical configurations is negligible, and the particles mainly 

659: vibrate around some classical equilibrium positions. In this case the elementary

660: excitations are phonons, and the quantization of the chain is reduced

661: to the quantization of phonon modes, see e.g. \cite{Hu01}. 

662: 

663: To single out low energy excitations we use the following approach.

664: Of course, a particular excitation can be selected if the corresponding mode

665: $V_{j}^{(l)}$ is known, but in general this is not a trivial task.

666: However, for low lying excitations one may expect that the modes have a simple harmonic

667: form\begin{eqnarray}

668: V_{j}(k_{l})\equiv V_{j}^{(l)} & = & \sqrt{\frac{2}{L}}\sin (k_{l}j),\label{eq:profV}\\

669:  &  & k_{l}=\pi l/L,\; (l=1,2,\ldots )\nonumber 

670: \end{eqnarray}

671: where the wavelength $\lambda _{l}\equiv 2\pi /k_{l}$ is

672: much larger than a characteristic size of inhomogeneity in the chain.

673: Direct numerical computations \cite{PhonFK} of phonon modes in the classical 

674: FK chain support this ansatz (\ref{eq:profV}).

675: 

676: %

677: \begin{figure}[ht]

678: \includegraphics[  clip,

679:   width=115mm,

680:   height=85mm,

681:   keepaspectratio,

682:   angle=90,

683:   origin=c]{fig2.ps}

684: 

685: \vskip -12mm

686: \caption{\label{fig:h08exmpla} The phonon spectral function $F^{(1)}(\omega _{m})$

687: versus the rescaled frequency $\omega _{m}$, data are shown

688: for the lowest spatial mode with $l=1$. The chain parameters are: 

689: $s/r=89/55$, $K=5$,$\tau _{0}=80$, $\hbar=0.8$. Here $m$ varies from $1$ to $89$,

690: but for clarity only selected values are shown. The solid

691: curve gives the fit by  Eqs.(\ref{eq:Fw},\ref{eq:Fw_renorm}), open circles show numerical

692: data. The fit determines the phonon frequency of the first mode ($\nu_1^2=3.170\pm0.012$)

693: and the renormalization factor $f=1.0034\pm0.0034$.}

694: \end{figure}

695: 

696: The numerical test of this anzatz (\ref{eq:profV}) is given in Fig.\ref{fig:h08exmpla}.

697: Here a typical result of quantum simulations of the spectral function

698: $F^{(l)}(\omega _{m})$ is shown for the lowest phonon

699: mode $l=1$ at $\hbar=0.8$. Fitting the data by Eq.(\ref{eq:Fw_renorm})

700: with the renormalization factor $f$ and the frequency of the

701: phonon mode $\nu _{l}$ as  free parameters, we obtain

702: $f=1.0034\pm 0.0034$ and $\nu_1^2=3.170\pm0.012$.

703: This fit shows that the renormalization factor $f$ is remarkably close

704: to unity, in spite of the fact that the value $\hbar =0.8$ is not

705: small. Hence, the ansatz (\ref{eq:profV})

706: provides a good approximation to actual profiles of lowest phonon

707: modes. This also indicates that the Gauss approximation used

708: in the Section \ref{sub:SpFunc} and in papers \cite{Hu98,Hu01r,Ho01}

709: works fine here. However we note that  at the same time the  quantum effects

710: renormalize substantially the phonon frequency ($\nu_1^2=3.170$)

711: compared to its value at $\hbar =0.2$  ($\nu _{1}^{2}=3.706\pm 0.019$).

712: The computations for the classical FK chain give $\nu _{1}^{2}=3.717$ ($\hbar=0$).

713: 

714: These data show that at small $\hbar$ the frequency $\nu_1$ obtained from

715: the quantum simulations approaches to the frequency of phonon mode in the classical chain.

716: This fact gives a very important check of the consistency of our quantum simulations. 

717: We stress that a good agreement between the numerical data 

718: of Fig.\ref{fig:h08exmpla} and the theoretical  spectral function (\ref{eq:Fw_renorm}) 

719: takes place in the \emph{whole} frequency range of $\omega _{m}$. 

720: This is, in fact, a very important consistency

721: check of the good relaxation of our paths ensemble at all frequencies,

722: including paths fluctuations at the lowest frequency available in

723: the system.

724: 

725: 

726: \begin{figure}[ht]

727: \includegraphics[ clip,

728:   width=115mm,

729:   height=80mm,

730:   keepaspectratio,

731:   angle=90,

732:   origin=c]{fig3.ps}

733: \vskip -10mm

734: \caption{\label{fig:h08exmplb} The phonon dispersion law $\nu (k)$: open circles show 

735: the data obtained from the fit as it is shown in Fig.\ref{fig:h08exmpla} for $l=1\div30$,

736: and the straight line shows the fit given by Eq.(\ref{eq:DispLaw}). The chain

737: parameters are the same as in Fig.\ref{fig:h08exmpla}, the wave vector $k=\pi l/L$.}

738: \end{figure}

739: 

740: By fitting the data for different phonon modes $l=1\div30$ one

741: can extract the dispersion relation for phonons  $\nu(k)$ 

742: where $k=\pi l/L$ (see Fig.\ref{fig:h08exmplb}). The majority of data

743: points follow the straight line given by the formula

744: \begin{equation}

745: \nu ^{2}(k)=\nu _{0}^{2}+c^{2}k^{2}

746: \label{eq:DispLaw}

747: \end{equation}

748: where $\nu _{0}$ is the phonon frequency gap, and $c$ is the velocity

749: of sound. The fit of numerical data gives: $\nu_0^2=3.204\pm 0.026$,

750: $c^2=15.0\pm 0.7$ for $\hbar=0.8$; $\nu_0^2=3.706\pm 0.019$,

751: $c^2=13.3\pm 0.5$ for $\hbar=0.2$. These quantum data should be compared

752: with the classical case where $\nu_0^2=3.697$, $c^2=11.5$ ($\hbar=0$).

753: We note that there is a difference between the frequency of the first

754: spatial harmonic $\nu_1$ and the frequency gap value $\nu_0$ obtained

755: from the dispersion law. However this difference is small and comparable

756: with the statistical errors. For small $\hbar$ the parameters of the 

757: dispersion law converge to their classical values. 

758: 

759: The described approach allows to obtain a complete information about 

760: low-energy phonon excitations in the whole quasiclassical region 

761: $\hbar \lesssim 1$.

762: 

763: 

764: \subsection{Transition region $1\lesssim \hbar \lesssim 2$.}

765: \label{sub:Transition-region}

766: 

767: As it is seen from Fig.\ref{fig:overview}, this region corresponds

768: to the transition between two regimes $\hbar \lesssim 1$ and 

769: $\hbar \gtrsim 2$ where the dependence of the quantum excitations on the particle

770: location in the chain looks quite regular.

771: 

772: As it will be shown later the irregular behavior in the region

773: $1\lesssim \hbar \lesssim 2$ is related to a significant increase

774: of the density of instantons. At high density the interaction between

775: instantons becomes important and results in onset of new phonon branch 

776: at $\hbar>2$.

777: In this section we discuss the properties of instantons and phonons 

778: and obtain estimates for their frequencies. 

779: 

780: Let us start with a discussion of tunneling effects. For $\hbar < 1$

781: the contribution of tunneling to the spectral function is exponentially

782: small being proportional to $\exp (-const/\hbar )$. However at  $\hbar >1$ 

783: the tunneling probability becomes large and it gives a significant contribution

784: to the spectral function. The transition to this regime is seen in  Fig.\ref{fig:overview} 

785: as a sequence of sharp isolated peaks. 

786: Following the pioneering paper \cite{Poly77}, a tunneling event can

787: be associated to an \emph{instanton}. In the imaginary time representation,

788: the instanton is a local jump between two wells, which is fast compared

789: to the mean time interval between subsequent jumps: while the size

790: of instantons (in time $\tau$) is practically independent of $\hbar $, the

791: separation between them is exponentially large in the quasiclassical

792: limit (low instanton density). In particular, this means that in the 

793: first approximation one may consider instantons as independent jumps, 

794: as can be also checked from a direct examination of our path ensemble.

795: Here we should stress two important properties of instantons

796: in the quantum FK chain:

797: 

798: (i) each jump of a particle $i$ in its 

799: position $x_i$ gives displacements of neighboring particles,

800: which decay exponentially with the distance from the jump

801: location, i.e. instantons are exponentially localized in space inside 

802: the chain;

803: 

804: (ii) instantons are distributed inhomogeneously along the chain since

805: the tunneling probability is highly sensitive to variations of barrier

806: heights due to chain inhomogeneity. 

807: 

808: A simple explanation of the exponential localization of an instanton

809: (along the chain) comes from the fact that the new static configuration

810: produced by it can be seen as a local \emph{static} defect on the

811: original configuration, which is known to dye away exponentially with

812: the classical Lyapunov exponent \cite{Au78,Au83b,Au83c}. Indeed,

813: in Fig.\ref{fig:overview} one can see that at $\hbar \sim 1.2\div 1.3$

814: instanton contributions are exponentially peaked around some particular

815: positions along the chain. 

816: 

817: Let us now consider the properties of elementary excitations originated

818: by instantons. There is a question how to select numerically a single instanton 

819: excitation.  Obviously, the ansatz (\ref{eq:profV}) used for phonons is good 

820: only for extended modes while instantons are localized in space. Therefore 

821: in this case we analyze numerically the frequency spectrum of a given

822: particle $i$ in the chain (defined by Eq.(\ref{eq:four_x})). If

823: $\hbar $ is not too high (close to one), instantons do not overlap, and the

824: main contribution to the spectrum comes from the instanton which is near to the given 

825: particle. This contribution reaches its maximum for a particle which actually

826: jumps.

827: 

828: At the same time besides instanton jumps the quantum motion of a particle 

829: in the chain contains a contribution of many phonons with different frequencies 

830: (see, e.g. Eq.(\ref{eq:Fw_x_i})). This phonon background should be subtracted in 

831: order to single out the contribution of instanton.  

832: Fortunately the frequencies $\nu _{l}$ of phonon excitations are much higher than 

833: the frequency $\nu ^{(inst)}$ of chosen instanton.

834: Hence these two types of excitations are well separated in the frequency domain.

835: Therefore in our analysis of the instanton contribution to the spectral function 

836: $F_i(\omega_m)\equiv\langle \left| a_{i}(\omega _{m})\right|^2\rangle$ 

837: we can restrict ourselves to the frequency domain $\omega _{m}<\omega _{bound}$,

838: where the boundary $\omega _{bound}$ is chosen by the condition 

839: $\nu ^{(inst)}\ll \omega _{bound}\ll \nu _{l}$

840: for all phonon modes $l$. In this frequency domain the phonon contribution

841: (\ref{eq:Fw}) is practically independent of $\omega _{m}$ and

842: can be replaced by some constant $C_{i}$. Thus we can extract

843: the frequency $\nu ^{(inst)}$ and the weight $f_{i}^{(inst)}$ of

844: the instanton excitation by following fit for the spectral function 

845: $F_{i}(\omega _{m})$:

846: 

847: \begin{equation}

848:  F_{i}(\omega _{m})=f_{i}^{(inst)}\hbar /\left(\omega _{m}^{2}+

849:  (\nu _{i}^{(inst)})^{2}\right)+C_{i}.\label{eq:fitInst}

850: \end{equation}

851: This fit contains three free parameters $\nu _{i}^{(inst)}$, $f_{i}^{(inst)}$ and

852: $C_{i}$. 

853: 

854: If instantons do not overlap then one may expect that

855: there are groups of particles which motion is dominated by a single

856: instanton. Inside each group the fit (\ref{eq:fitInst})

857: should give the same values for the frequencies $\nu _{i}^{(inst)}$, 

858: while the variation of weight $f_{i}^{(inst)}$ with $i$ 

859: determines the instanton profile along the

860: chain. This case is illustrated in Fig.\ref{fig:xEinst}a, which

861: corresponds to the early onset of instanton contribution at $\hbar =1.2$.

862: The six peaks in the bottom part of Fig.\ref{fig:xEinst}a show six non 

863: overlapping instantons, while the top part shows the corresponding frequencies 

864: as a function of particle index $i$ inside the chain. The peaks have different 

865: amplitudes, and the highest three of them involve groups of three particles

866: which have the same frequency inside each group. 

867: 

868: At higher $\hbar $

869: the number of instantons starts to grow rapidly and they begin to overlap.

870: A direct confirmation of this trend is seen in Fig.\ref{fig:xEinst}b

871: which corresponds to $\hbar =1.8$. Here all instantons have about ten percent

872: overlap with their neighbors and their interaction is rather

873: strong. As a result the step-like structure of frequencies, seen at

874: the top of the Fig.\ref{fig:xEinst}a, is practically destroyed. Thus

875: the instantons are ``collectivized'' and the consideration

876: of a single instanton as one-particle jump over barrier becomes 

877: not adequate. At higher $\hbar$ values this process leads to appearance

878: of a new phonon mode.  

879: %

880: \begin{figure}[ht]

881: \includegraphics[  width=115mm,

882:   height=85mm,

883:   keepaspectratio,

884:   angle=90,

885:   origin=c]{fig4.ps}

886: \vskip-12mm

887: \caption{\label{fig:xEinst}Dependence of the instanton frequency 

888: $\nu _{i}^{(inst)}$ (top)

889: and its weight $f_{i}^{(inst)}$ (bottom) on the position $i$ inside the chain 

890: (see text for explanations). (a) $\hbar=1.2$. Instantons do not overlap: 

891: the three highest peaks involves a group of three particles each. 

892: (b) $\hbar=1.8$ Instantons overlap. Chain parameters are

893: $s/r=34/21$, $K=5$, $\tau _{0}=320$.}

894: \end{figure}

895: 

896: 

897: %

898: \begin{figure}[th]

899: \includegraphics[  width=115mm,

900:   height=85mm, 

901:   keepaspectratio,

902:   angle=90,

903:   origin=c]{fig5.ps}

904: \vskip -12mm

905: \caption{\label{fig:spIP}Rescaled spectral function 

906: $F^{R}(\omega ,k_{1})=(\nu _{1}^{2}/\hbar )F(\omega ,k_{1})$

907: at the very beginning of instanton onset. The bump which appears at 

908: $\omega\lesssim 0.2$ corresponds to instanton admixture to the phonon 

909: spectral function. Black points, open circles and stars

910: correspond to $\hbar=1,1.1$ and $1.2$.

911: Lines show the fit to numerical data with Eq.(\ref{eq:fitPhIn}). Chain parameters

912: used in simulations are: $s/r=34/21$, $K=5$,$\tau _{0}=320$.}

913: \end{figure}

914: 

915: 

916: Let us now discuss the phonon properties in the region

917: $1\lesssim \hbar \lesssim 2$. As in the previous Section \ref{sub:Quasiclassical},

918: we extract them from the the spectral

919: function $F(\omega _{m},k_l)\equiv F^{(l)}(\omega _{m})$ obtained on the basis 

920: of the ansatz (\ref{eq:profV}) for a phonon mode $l$.  

921: However, in this region of $\hbar $

922: the spectral function $F^{(l)}(\omega _{m})$ has an admixture of instantons,

923: which grows rapidly with $\hbar $: a change

924: of $\hbar $ from $1.1$ to $1.2$ results in more than ten times

925: of the admixture weight(see Fig.\ref{fig:spIP}). We note that 

926: in the absence of instanton contribution the rescaled 

927: phonon spectral function 

928: $F^{R}(\omega ,k_{l})\equiv(\nu _{l}^{2}/\hbar )F(\omega ,k_{l})$

929: plotted in Fig.\ref{fig:spIP} should have an

930: universal limit equal to unity independent of the value of $\hbar$.

931: Hence the increase of $F^{R}(\omega ,k_{l})$ at small $\omega$

932: stresses the important contribution of instantons.

933: These instantons have different frequencies and their contribution

934: to the spectral function can be rather complicated.  

935: However, we can use again the strong frequency separation between instanon 

936: and phonon excitations. Indeed, for $\omega _{m}\gtrsim \nu _{l}\gg \nu ^{(inst)}$, 

937: all instanton contributions have an universal behavior  

938: $\propto \omega _{m}^{-2}$. Therefore we may replace them by a single 

939: ``instanton contribution'' with some average instanton frequency $\bar{\nu}^{inst}$. 

940: Then the spectral function can be fitted by a sum of two contributions:

941: \begin{eqnarray}

942:   F(\omega_{m},k_{l}) & = & f_{ph}\hbar /\left(\omega_{m}^{2}+

943:   \nu_l^{2}(k_{l})\right)\nonumber \\

944:    &  & +f_{inst}\hbar /\left(\omega_{m}^{2}+(\bar{\nu}^{inst}(k_{l})^2)\right)

945: \label{eq:fitPhIn}

946: \end{eqnarray}

947: where $f_{ph}(k),\nu_l(k)$ and $f_{inst}(k),\bar{\nu}^{in}(k)$ are

948: free fit parameters for phonons and instantons, respectively. 

949: 

950: Fitting the data for different phonon modes at $l=1-30$ we extract

951: the phonon dispersion law (\ref{eq:DispLaw}) (see Fig.\ref{fig:DLinTR},

952: compare with fit procedure for $\hbar<1$).

953: Contrary to the case of Fig.\ref{fig:h08exmplb} at $\hbar<1$, now

954: the data for the dispersion law $\nu (k)$ are scattered inside  

955: some finite band.

956: This indicates that the anzatz (\ref{eq:profV})

957: for the phonon profile is not so good to single out particular phonon

958: modes. Actually, the width of the

959: band provides some measure of the inaccuracy. Nevertheless, the phonon

960: modes are still approximately defined in the domain of $\hbar $ under consideration.

961: Their frequency decreases as $\hbar \to 2$ but remains separated

962: from zero by a finite gap.

963: 

964: %

965: \begin{figure}[th]

966: \includegraphics[  clip,

967:   width=110mm,

968:   height=85mm,

969:   keepaspectratio,

970:   angle=90,

971:   origin=c]{fig6.ps}

972: \vskip -10mm

973: \caption{\label{fig:DLinTR} Frequency $\nu (k)$ of phonons

974: versus wavenumber $k$ obtained from the fit (\ref{eq:fitPhIn})

975: at $\hbar =1.5$ (in the middle of the transition region).

976: Chain parameters used in simulations

977: are: $s/r=89/55$, $K=5$,$\tau _{0}=80$.}

978: \end{figure}

979: 

980: We also note that the fit (\ref{eq:fitPhIn}) allows formally

981: to determine the dispersion law $\nu _{inst}(k)$ for the instanton branch.

982: However the numerical data give irregular scattering of points inside a band

983: $0 \leq \nu_{inst}^2 \lesssim 0.01$ without any clear dependence on $k$. 

984: %(see Fig.\ref{fig:DLinTR}b).

985: The reason of such behavior is simple: projections of irregular positions of instantons

986: on the harmonic ansatz (\ref{eq:profV}) produce random weights for

987: contributions of different instantons. This result represents another 

988: manifestation of glass-like structure formed by instantons frozen/pinned

989: inside the chain. Since the positions of instantons are random the phonons

990: cannot propagate along the chain on large distances. In fact they become

991: localized by disorder in a way similar to the one-dimensional Anderson localization

992: (more details on the phonon properties in this regime will be presented elsewhere

993: \cite{PhonFK}).  

994: 

995: 

996: 

997: \subsection{New sliding phonon branch at $\hbar >2$}

998: \label{sub:New-phon}

999: 

1000: \begin{figure}[ht]

1001: \includegraphics[  width=115mm,

1002:   height=85mm,

1003:   keepaspectratio,

1004:   angle=90,

1005:   origin=c]{fig7.ps}

1006: \caption{\label{fig:path23} A sample of quantum paths of particles inside 

1007: some chain fragment, which corresponds to periods of the external potential 

1008: with numbers $22\div30$. Dashed lines show bottoms of the wells, thick solid

1009: lines show the tops of the barriers.

1010: Note an example of highly correlated instanton transitions at $\tau=15-30$

1011: which involve particles  in up to 5 periods of the potential.

1012: Parameters of the simulation are: $\hbar=2.3$, $s/r=89/55$, $K=5$ and 

1013: $\tau _{0}=80$. }

1014: \end{figure}

1015: 

1016: From Fig.\ref{fig:overview} one can see that the variation 

1017: of the amplitude $a_{i}(\omega_1)$ with $i$ (low frequency excitations)

1018: becomes rather smooth at $\hbar> 2$. 

1019: This means that the instantons are strongly overlapped here.

1020: In fact, this means 

1021: that several particles of some chain fragment jumps from one well to the 

1022: next simultaneously (see Fig.\ref{fig:path23}). 

1023: This is nothing but sliding of a local chain fragment

1024: along the periodical potential. The typical size of such fragments should

1025: grow with $\hbar $. If their sizes reach the size of the chain  then the sliding

1026: mode becomes open, and the phonon gap disappears. At this point the pinned

1027: instanton glass turns into the sliding phonon gas.

1028: 

1029: %

1030: \begin{figure}[th]

1031: \includegraphics[  width=115mm,

1032:   height=85mm,

1033:   keepaspectratio,

1034:   angle=90,

1035:   origin=c]{fig8.ps}

1036: \vskip -10mm

1037: \caption{\label{fig:DLinSR} The dispersion law $\nu (k)$ for sliding 

1038: phonons at $\hbar =2.5$. The numerical data (circles with error bars)

1039: are obtained from the fit (\ref{eq:fitPhIn}) for the instanton branch

1040: $\nu_{inst}(k_l)$.

1041: The straight line shows the best fit (\ref{eq:DispLaw}) to numerical data. The chain

1042: parameters used in simulations are: $s/r=89/55$, $K=5$,$\tau _{0}=80$.}

1043: \end{figure}

1044: 

1045: 

1046: A confirmation of this picture is presented in Fig.\ref{fig:DLinSR},

1047: which corresponds to $\hbar =2.5$ being just above the transition point

1048: $\hbar_c\approx 2$. The numerical data for the dispersion law $\nu (k)$

1049: in Fig.\ref{fig:DLinSR} are obtained from the fit (\ref{eq:fitPhIn}) 

1050: at $\hbar >2$. Here the behavior of phonon and instanton modes changes 

1051: dramatically: phonons data $\nu (k)$ are now irregularly scattered over a wide 

1052: band while the data points for instanton branch follow a single line, 

1053: reproducing fairly well a \emph{phonon}-like dispersion law with  zero gap

1054: (see Figs.\ref{fig:DLinSR},\ref{fig:Esurf}).

1055: In particular, the fit (\ref{eq:DispLaw}) gives 

1056: $\nu _{0}=0.04\pm 0.01$, which is close to zero. In fact, 

1057: this value is smaller 

1058: than the minimal frequency $2\pi /\tau _{0}$ ($\approx 0.079$, 

1059: at $\tau _{0}=80$) and therefore it is compatible with zero. 

1060: 

1061: On the contrary, the wide scattering of data points for the phonon

1062: branch indicates that the ansatz (\ref{eq:profV}) is not good for phonon 

1063: contribution at $\hbar\gtrsim 2$ (see Fig.\ref{fig:Esurf}). This scattering of points

1064: is related to the localization of high-frequency

1065: phonon modes. In contrast, a smooth behavior of data points for the instanton

1066: branch demonstrates that the instanton wave functions become close

1067: to the harmonic wave ansatz (\ref{eq:profV}). Hence these excitations

1068: are delocalized. This leads us to the conclusion, that 

1069: for $\hbar>2$ the instanton branch is replaced by a new gapless

1070: branch of new sliding phonons. 

1071: 

1072: 

1073: \subsection{Global picture of elementary excitations in the FK chain}

1074: \label{sub:SummaryExcit}

1075: 

1076: The ensemble of data for the dispersion law $\nu(k)$ of elementary excitations

1077: at different values of $\hbar$ is shown in Fig.\ref{fig:Esurf} by the two sheets

1078: representing the phonon branch (1) and the instanton branch (2). 

1079: The numerical data are obtained with the ansatz (\ref{eq:profV}) by the fit 

1080: (\ref{eq:fitPhIn}). 

1081: %

1082: \begin{figure}[ht]

1083: \includegraphics[clip,

1084:   width=120mm,

1085:   height=88mm,

1086:   keepaspectratio,

1087:   angle=270,

1088:   origin=c]{fig9.ps}

1089: \vskip -15mm

1090: \caption{\label{fig:Esurf} The frequency $\nu (k)$ of elementary excitations

1091: versus the wavenumber $k$ at different $\hbar $. Chain parameters used

1092: in simulations are: $s/r=89/55$, $K=5$,$\tau _{0}=80$. The two sheets (1) and (2)

1093: refer to phonon and instanton excitations respectively.} 

1094: \end{figure}

1095: 

1096: 

1097: The sheet (1) refers to phonons originated from classical phonon

1098: modes which are well reproduced in the limit $\hbar \to 0$. The frequencies

1099: of these modes are well separated from zero by a large gap.  Therefore

1100: one may say that  they  form the optical phonon branch. At

1101: $\hbar \gtrsim 1$ these modes show a tendency to become softer, and at 

1102: $\hbar \gtrsim 1.5$ their dependence on $k$ and $\hbar $ becomes irregular.

1103: As it was explained in the previous section this irregular dependence 

1104: is related to the glass-like structure of randomly pinned instantons

1105: which density increases with the growth of $\hbar$.

1106: The sheet (2) appears at $\hbar <2$ from the instanton contributions

1107: into the Feynman path integral. For $\hbar>\hbar_c\approx 2$

1108: this instanton branch turns into a new gapless branch of sliding phonons

1109: (see the discussion related to Fig.\ref{fig:DLinSR}).

1110: 

1111: %

1112: \begin{figure}[h]

1113: \includegraphics[  width=115mm,

1114:   height=85mm,

1115:   keepaspectratio,

1116:   angle=90,

1117:   origin=c]{fig10.ps}

1118: 

1119: \caption{\label{fig:PhSummary} Dependence of the phonon gap $\nu _{0}$ 

1120: and sound 

1121: velocity $c$ on $\hbar $ for the case of Fig.\ref{fig:Esurf} at $K=5$.

1122: Open symbols correspond to the sheet (1) in Fig.\ref{fig:Esurf}.

1123: Circles and triangles are obtained at $\tau_0=80$ for chain sizes 

1124: $s/r=89/55$ and $233/144$ respectively. Stars correspond

1125: to $s/r=34/21$ and $\tau _{0}=320$. The solid

1126: line (bottom) gives the fit (\ref{eq:CritExpC2}) for  $1.3 \leq\hbar\leq 2$

1127: (see text). The full circles (top) for $\hbar>2$ refer to the sliding 

1128: phonon branch from the sheet(2) of Fig.\ref{fig:Esurf}, they 

1129: indicate a zero phonon gap.}

1130: \end{figure}

1131: 

1132: 

1133: A more quantitative picture can be obtained from the

1134: numerical data for the gap $\nu _{0}$ and the sound velocity $c$ for 

1135: both sheets in Fig.\ref{fig:Esurf}.

1136: For the optical phonon branch the values of $\nu_0$ and $c$ are determined

1137: from the fit (\ref{eq:fitPhIn}) for different values of $\hbar$ 

1138: (see Fig.\ref{fig:PhSummary}).  

1139: The data show that the phonon gap remains finite and 

1140: large at $\hbar<\hbar_c\approx 2$. 

1141: In contrast, the sound velocity $c$ drops to zero as $\hbar$ 

1142: approaches the value $\hbar_c$. This decay is compatible with

1143: the fit 

1144: \begin{equation}

1145:   c^{2}=a(\hbar _{c}-\hbar )^{\alpha }

1146: \label{eq:CritExpC2}

1147: \end{equation}

1148: shown by the solid line with $a=20.5\pm 1.3$, $\hbar _{c}=2.0\pm 0.1$

1149: and the critical exponent $\alpha =0.52\pm 0.07$.

1150: Even if the numerical data for $c$ have certain fluctuations they still 

1151: clearly indicate the quantum phase transition at $\hbar_c\approx 2$,

1152: where the sound velocity $c$ drops to zero. Further extensive numerical studies 

1153: are required to determine the behavior in the vicinity of transition

1154: in a more precise way.

1155: %

1156: \begin{figure}[h]

1157: \includegraphics[  width=115mm,

1158:   height=85mm,

1159:   keepaspectratio,

1160:   angle=90,

1161:   origin=c]{fig11.ps}

1162: \caption{\label{fig:InSummary} Dependence of sound velocity $c$

1163: on $\hbar $ for excitations on the soft phonon sheet (2) 

1164: in Fig.\ref{fig:Esurf}. Points correspond to chain parameters $s/r=89/55$ 

1165: $K=5$ and $\tau _{0}=80$. The solid line shows the linear fit to the

1166: sound velocity data inside the region $\hbar >2$ (see text).}

1167: \end{figure}

1168: 

1169: 

1170: For the sliding phonon branch  we extract the parameters $\nu_0$ and $c$ from the data of 

1171: sheet (2) in Fig.\ref{fig:Esurf} using a more general fit 

1172: given by 

1173: \[

1174:   \nu ^{2}(k)=\nu _{0}^{2}+c^{2}k^{2}/(1+k^{2}/k_{B}^{2}).

1175: \]

1176: Compared to the standard case (\ref{eq:DispLaw}) 

1177:  we introduce an additional parameter $k_B$

1178: to take into account the saturation of $\nu(k)$ at large $k$ (see, 

1179: Fig.\ref{fig:DLinSR},\ref{fig:Esurf}).

1180: The numerical data show that the gap 

1181: $\nu _{0}$ is small and does not exceed the minimal frequency in 

1182: the system $\varpi =2\pi /\tau _{0}$. Hence, the gap $\nu_0$ is compatible with zero

1183: (see Fig.\ref{fig:PhSummary} (top)). 

1184: At the same time  

1185: the sound velocity $c$ for the sliding phonon branch grows approximately

1186: linearly with $\hbar $ (see Fig.\ref{fig:InSummary}). 

1187: The best fit of numerical data for $\hbar>2$ gives 

1188: \begin{equation}

1189:   c^{2}=a\cdot (\hbar -b),\label{eq:C2ofHbar}

1190: \end{equation}

1191: with $a=7.6\pm 0.1$, $b=1.57\pm 0.05$. Formally, the value of $b$ is different

1192: from the value $\hbar_c=2.0$ in (\ref{eq:CritExpC2}). However in view of large statistical

1193: fluctuations both fits for optical phonons and sliding phonons are compatible with the 

1194: quantum phase transition at $\hbar_c\approx 2$.

1195: 

1196: \subsection{Time correlations}

1197: \label{sub:Time-corr}

1198: 

1199: Time correlations are closely related to the frequency spectrum of

1200: elementary excitations in the system. Their analysis is probably the

1201: most traditional way to extract properties of the elementary excitations,

1202: see, e.g. \cite{Creu81,Shur84}. Below we discuss the connection

1203: of this traditional method with 

1204: our approach based on the Fourier spectrum of Feynman paths

1205: described above. We compare the results obtained by these two different methods.

1206: 

1207: The one-particle time correlators are directly related to the Fourier spectrum

1208: of Feynman paths, and are given by the following expression:

1209: \[

1210:  \left\langle x_{i}(0)x_{i}(\tau )\right\rangle =

1211:  \frac{1}{\sqrt{\tau _{0}}}\sum _{\omega _{m}}

1212:  \left|a_{i}(\omega _{m})\right|^{2}\exp \left(-i\omega _{m}\tau \right)

1213: \]

1214: In fact, many elementary excitations with different frequencies contribute 

1215: to a particle motion in the chain. In order to

1216: single out a particular phonon mode we study the correlators of

1217: normal modes $X^{l}(\tau )=\sum _{i}V_{i}^{(l)}x_{i}(\tau )$

1218: where $V_i^{(l)}$ are eigenvectors defined in Eqs.(\ref{eq:Om2})-(\ref{eq:Om3}).

1219: Then from Eq.(\ref{eq:corrAA}) we obtain

1220: \begin{eqnarray}

1221:  \left\langle X^{l}(0)X^{l}(\tau )\right\rangle  &  & =

1222:  \frac{1}{\sqrt{\tau _{0}}}\sum _{\omega _{m}}\left|A^{l}(\omega _{m})\right|^{2}

1223:  \exp \left(-i\omega _{m}\tau \right)\nonumber \\

1224:  = &  & \left\langle \left(X^{l}(0)\right)^{2}\right\rangle 

1225:  \left(\textrm{e}^{-\nu _{l}\tau }+

1226:  \textrm{e}^{-\nu _{l}(\tau _{0}-\tau)}\right),

1227: \label{eq:CorrPhT}

1228: \end{eqnarray}

1229: where $\left\langle \left(X^{l}(0)\right)^{2}\right\rangle =\hbar /2\nu _{l}$

1230: is the contribution of a single phonon mode and the periodicity along

1231: the time torus results in a second exponential term in (\ref{eq:CorrPhT}). 

1232: 

1233: However, Eq.(\ref{eq:corrAA}) assumes only small 

1234: quantum fluctuations ($\sim \hbar $) around some \emph{classical} trajectory. 

1235: But due to tunneling effects (or instantons) a particle jumps from one

1236: \emph{classical} trajectory to another and its actual motion

1237: is given by a sum of phonon $x_{i}^{(ph)}(\tau )$

1238: and instanton $x_{i}^{(inst)}(\tau )$ contributions:

1239: \[

1240:   x_{i}(\tau )=x_{i}^{(ph)}(\tau )+x_{i}^{(inst)}(\tau ).

1241: \]

1242: In general, both motions influence each other, but in the quasiclassical

1243: limit $\hbar \to 0$ they can be considered as independent. In this limit they

1244: have quite different frequency scales: the phonon

1245: frequency $\nu_l$ is of the order of $K^{1/2}$ while the frequency of tunneling jumps

1246: $\nu_{inst}$

1247: is exponentially small. In contrast, while the amplitude of phonon

1248: oscillations in the limit $\hbar \to 0$ is small: 

1249: $\left\langle \left(x_{i}^{(ph)}(\tau )\right)^{2}\right\rangle \propto \hbar $,

1250: the amplitude of jumps is defined by difference between equilibrium

1251: particle positions in two neighbor wells (in our case it is $\sim 3\div 4$),

1252: i.e. it does not depend on $\hbar $. Hence, for a jumping particle

1253: $\left\langle \left(x_{i}^{(inst)}(\tau )\right)^{2}\right\rangle \sim 10$

1254: which is not small even in the quasiclassical limit $\hbar \to 0$. 

1255: Therefore the instanton contribution to the time correlator 

1256: has the form \cite{Shur84}

1257: %\[

1258: % \left\langle x_{i}(0)x_{i}(\tau )\right\rangle _{inst}=

1259: % \left\langle \left(x_{i}^{(inst)}(\tau )\right)^{2}\right\rangle \times \\

1260: % \left(\textrm{e}^{-\nu _{inst}\tau }+ 

1261: % \textrm{e}^{-\nu _{inst}(\tau _{0}-\tau )}\right).

1262: %\]

1263: \begin{eqnarray}

1264:  \left\langle x_{i}(0)x_{i}(\tau )\right\rangle _{inst}=

1265:  \left\langle \left(x_{i}^{(inst)}(\tau )\right)^{2}\right\rangle \times \\

1266:  \left(\textrm{e}^{-\nu _{inst}\tau }+ 

1267:  \textrm{e}^{-\nu _{inst}(\tau _{0}-\tau )}\right). \nonumber

1268: \end{eqnarray}

1269: The pre-exponent factor is large for jumping particles even in the deep quasiclassical 

1270: regime.

1271: 

1272: In fact, not any particle can easily jump from one well to another: different

1273: classical trajectories have different actions and all jumps which

1274: result in a large change of action $\Delta S\gtrsim \hbar $ are

1275: inhibited. In particular, it is clearly seen in Fig.\ref{fig:overview}

1276: that the number of instanton peaks is smaller at smaller $\hbar$ since

1277: the contribution of transitions with large difference in action $\Delta S$

1278: is suppressed.

1279: 

1280: Finally, the general form for the time correlator 

1281: which takes into account the instantons contribution takes the form:

1282: \begin{eqnarray}

1283:  C(\tau ) & \equiv  & \left\langle X^{l}(0)X^{l}(\tau )\right\rangle =\nonumber \\

1284:  & = & \left\langle \left(X^{l}(0)\right)^{2}\right\rangle 

1285:  \left(\textrm{e}^{-\nu _{l}\tau }+\textrm{e}^{-\nu _{l}(\tau _{0}-\tau )}\right)

1286: \nonumber\\

1287:  & + & \sum _{inst}w_{in}\left(\textrm{e}^{-\nu _{inst}\tau }+

1288:  \textrm{e}^{-\nu _{inst}(\tau _{0}-\tau )}\right).

1289: \label{eq:CorrPhInT}

1290: \end{eqnarray}

1291: Here the first term describes the phonon contribution and in the second term

1292:  the sum is taken over instantons and the instanton weights 

1293: $w_{inst}=\sum _{i,k}V_{i}^{(l)}V_{k}^{(l)}

1294:  \left\langle x_{i}(0)x_{k}(0)\right\rangle $

1295: describe the overlap with the ansatz (\ref{eq:profV}). Both types 

1296: of contributions (\ref{eq:CorrPhInT})

1297: are clearly seen in Fig.\ref{fig:CorrTa} at $\hbar =1.4$. 

1298: The initial rapid drop at $\tau\lesssim 1.5$ 

1299: corresponds to phonon contribution while the slow decay

1300: at $\tau>1.5$ corresponds to the instanton contribution. 

1301: This initial drop is related to the existence of large

1302: quasiclassical gap for phonon excitations.

1303: For $\hbar=2.5>\hbar_c$ the gap disappears and the correlator decay

1304: very slowly (see Fig.\ref{fig:CorrTa}).

1305: \begin{figure}[h]

1306: \includegraphics[clip,

1307:   width=115mm,

1308:   height=85mm,

1309:   keepaspectratio,

1310:   angle=90,

1311:   origin=c]{fig12.ps}

1312: \caption{\label{fig:CorrTa} The numerically computed time correlator 

1313: $C(\tau )=\left\langle X^l(0)X^l(\tau )\right\rangle$ for $l=1$ at

1314: different time separations $\tau $ for $\hbar =1.4$ (lower points) and 

1315: $\hbar=2.5$ (upper points). The parameters of the chain are $s/r=89/55$ 

1316: $K=5$ and $\tau _{0}=80$.

1317: }

1318: \end{figure}

1319: 

1320: For $\hbar>\hbar_c$ we have a new phase where

1321: instantons are replaced by sliding phonons.

1322: Therefore in this regime we fit the numerical data for $C(\tau)$

1323: by Eq.(\ref{eq:CorrPhInT}) with $w_{inst}=0$. The results for 

1324: different $l$ allow to obtain numerically the dispersion law

1325: $\nu(k)$ shown in Fig.\ref{fig:CorrTb} (open circles). However

1326: the accuracy of this data is not so good compared to data

1327: (full circles) obtained 

1328: from the analysis of Fourier spectrum of Feynman paths described in the 

1329: previous sections.

1330:  

1331: \begin{figure}[h]

1332: \includegraphics[clip,

1333:   width=115mm,

1334:   height=85mm,

1335:   keepaspectratio,

1336:   angle=90,

1337:   origin=c]{fig13.ps}

1338: \caption{\label{fig:CorrTb} Frequency of elementary excitations $\nu (k)$ 

1339: versus $k$ for sliding phonon branch at $\hbar =2.5$. Open circles show 

1340: data extracted from the  time correlator $C(\tau)$ for different $l$ 

1341: (see Fig.\ref{fig:CorrTa}),

1342: full circles present the results obtained from the fit (\ref{eq:fitPhIn})

1343: for the frequency spectrum of Feynman paths (see Fig.\ref{fig:DLinSR}). 

1344: The chain parameters are the same as in Fig.\ref{fig:CorrTa}.

1345: }

1346: \end{figure}

1347: 

1348: 

1349: 

1350: \section{quantum phase transition}

1351: \label{sec:QFT}

1352: 

1353: The structual rearrangement of the elementary excitations spectrum can

1354: be related to \emph{quantum phase transition} in the chain from a

1355: ``pinned'' to ``sliding'' phase. However, we would like to stress that

1356: in contrast to the classical picture \cite{Au83b,Au83c} in the

1357: quantum case the absence of energy gap for excitations is not 

1358: necessarily related to the opening of the sliding phase. Indeed,

1359: due to quantum tunneling through Peierls-Nabbarro barriers related to

1360: instantons there are excitations

1361: with energy which decreases exponentially with the increase of barrier heights. 

1362: Formally, this corresponds to the disappearance of excitation gap. 

1363: Therefore, to confirm firmly

1364: the appearance of sliding phase we need to consider spatial correlations

1365: of particle motion in the chain. The sliding phase appears when

1366: the spatial correlation 

1367: length becomes comparable with the length of the chain.

1368: 

1369: 

1370: \subsection{Spatial correlation length}\label{sec:CorrLen}

1371: 

1372: The analysis of spatial correlations (correlations between the motion

1373: of different particles in the chain) is the most evident way to observe

1374: the transition between pinned and sliding phases. In principle, the

1375: spatial

1376: correlation function can be explicitly computed if the spectrum

1377: and the wave functions of elementary excitations are known. In fact, we

1378: have a complete quantitative picture for \emph{phonon} modes, at least

1379: for the low-lying ones. To obtain numerically the value of the spatial

1380: correlation length $l_c$ we assume that the elementary excitation

1381: spectrum is given by the dispersion relation 

1382: $\nu ^{2}(k)=\nu _{0}^{2}+c^{2}k^{2}$ 

1383: and the corresponding phonon modes are given the ansatz 

1384: (\ref{eq:profV}). Then the same-time spatial correlator reads:

1385: \begin{eqnarray}

1386:  \left\langle \left(x_{i}-\left\langle x_{i}\right\rangle \right)\left(x_{j}-

1387:  \left\langle x_{j}\right\rangle \right)\right\rangle  & = & 

1388:  \sum _{l}V_{i}^{(l)}V_{k}^{(l)}

1389:  \left\langle \left(X^{l}(0)\right)^{2}\right\rangle \nonumber \\

1390:  =\hbar \sum _{k}\frac{\cos (k(i-j))}{L\nu (k)} & = & 

1391:   \frac{\hbar }{\pi c}K_{0}\left(\left|i-j\right|\right/l_c),

1392: \label{eq:CorrX}

1393: \end{eqnarray}

1394: where $K_{0}(x)$ is McDonald's function, with a known asymptotics:

1395: $K_{0}(x)\rightarrow _{x\rightarrow \infty }\sqrt{\frac{\pi }{2x}}\textrm{e}^{-x}$,

1396: and $l_c= c/\nu _{0}$. 

1397: 

1398: 

1399: The fit (\ref{eq:CorrX}) of the numerical data gives the value

1400: of $l_c$  for different values of $\hbar $ as it is

1401: shown in Fig.\ref{fig:CorrX}. It is seen that the length $l_c$ has

1402: a sharp increase at $\hbar_c \approx 2$, and for $\hbar>\hbar_c$

1403: it becomes comparable with the length of the chain.

1404: This indicates that we have the \emph{quantum phase transition}

1405: near $\hbar_c\approx 2$.

1406: Indeed at $\hbar <2$ the length $l_c$

1407: is practically independent of the chain length $L$ while at $\hbar >2$

1408: it starts to increase with $L$. This is confirmed by the data of

1409: Fig.\ref{fig:CorrX} where in spite of strong fluctuations for $\hbar>2$

1410: the length $l_c$ becomes comparable with the chain size $L$.

1411:  

1412: \begin{figure}[H]

1413: \includegraphics[  width=65mm,

1414:   height=85mm,

1415:   angle=90,

1416:   origin=c]{fig14.ps}

1417: \caption{\label{fig:CorrX}The dependence of the spatial correlation length 

1418: $l_{c}$ on $\hbar $. Crosses,

1419: full and open circles correspond to chains with $s/r=34/21$,

1420: $89/55$ and $233/144$, respectively; $K=5$, $\tau _{0}=80$. 

1421: The vertical dot-dashed line marks the quantum phase transition

1422: at $\hbar_c\approx 2$.

1423: The positions of the horizontal dotted lines are proportional to the chain length 

1424: $L=2\pi r$.

1425: }

1426: \end{figure}

1427: 

1428: \subsection{Longwave response}

1429: 

1430: The appearance of the new sliding phonon phase implies

1431: that the response of amplitude of the longwave modes should be large

1432: in this regime. To test this expectation we present the dependence

1433: of the amplitude $A^l(\omega=0)$ on $\hbar$ in Fig.\ref{fig:X0sk}.

1434: The numerical data demonstrate a sharp increase of $A^l(0)$

1435: near $\hbar_c\approx 2$. It is interesting to note that due to the existence

1436: of frequency gap for phonon excitations at $\hbar<\hbar_c$ the amplitude $A^l(0)$

1437: is not very sensitive to the variations of $l$. On the contrary, for 

1438: $\hbar>\hbar_c$ the gap disappears and $A^l(0)$ starts to depend on $l$ 

1439: (see Fig.\ref{fig:X0sk}a). In a similar way $A^l(0)$ is independent of

1440: the chain length $L$ for $\hbar<\hbar_c$ while at $\hbar>\hbar_c$

1441: it grows with $L$ (see Fig.\ref{fig:X0sk}b).

1442: The numerical data of Fig.\ref{fig:X0sk}b for $1.5\leq\hbar\leq 2$

1443: can be described by the fit

1444: \begin{equation}\label{eq:gammaA2}

1445:   \langle (A^{l}(0))^2\rangle\approx A (\hbar_c-\hbar)^{-\gamma} 

1446: \end{equation} 

1447: which gives $\gamma=5.06\pm 1.72$ and $\hbar_c=2.01\pm0.05$. 

1448: 

1449: %

1450: \begin{figure}[H]

1451: \includegraphics[  width=115mm,

1452:   height=85mm,

1453:   keepaspectratio,

1454:   angle=90,

1455:   origin=c]{fig15.ps}

1456: \caption{\label{fig:X0sk}The amplitude square of zero-frequency quantum fluctuations

1457: $(A^{l}(0))^2$ as a function of $\hbar$ for $K=5$ and $\tau _{0}=80$. 

1458: (a) Full circles, squares

1459: and open circles are for modes $l=1,3$ and $5$, respectively,

1460: for the chain with $s/r=89/55$. (b) Open circles, full circles

1461: and stars are for $s/r=233/144$, $89/55$ and $34/21$, respectively;

1462: $l=1$. The vertical dot-dashed lines mark $\hbar_c\approx 2$,

1463: the horizontal dotted lines give the average values for $\hbar>2$. 

1464: }

1465: \end{figure}

1466: 

1467: \subsection{Other characteristics}

1468: 

1469: 

1470: Another way to test the transition from the pinned instanton glass to the sliding phonon phase

1471: is to measure the sensitivity to small shifts of boundaries of the chain. With this aim we consider the shift of boundary particles $i=0$ and $i=s$ given by 

1472: \begin{equation}

1473:    x_0(\tau)=a_S \cos(2 \pi\tau/\tau_0), \;\; x_s(\tau)=x_0+L,

1474: \end{equation} 

1475: where the amplitude $a_S$ was fixed at $a_S=0.5$. In Fig.\ref{fig:slide} we present the

1476: dependence of the response function 

1477: $R(i)=\langle x_0 (x_i-\langle x_i\rangle\rangle/\langle x_0^2 \rangle$

1478: on the particle number $i$ inside the chain.

1479: It is seen that the response in the center of the chain 

1480: drops strongly when the parameter $\hbar$ changes from 

1481: $\hbar=2.2$  to $\hbar=1.8$. This means that the chain is locked for 

1482: $\hbar<\hbar_c\approx 2$ while for $\hbar>\hbar_c$ the chain

1483: slides following the displacements of the boundary

1484: particles.

1485: \begin{figure}[ht]

1486: \includegraphics[  width=115mm,

1487:   height=85mm,

1488:   keepaspectratio,

1489:   angle=90,

1490:   origin=c]{fig16.ps}

1491: \caption{\label{fig:slide} Dependence of the response function $R(i)$ on 

1492: the particle position $i$

1493: in the chain at $\hbar=1.8$ (full circles) and $\hbar=2.2$ (open circles).

1494: The horizontal dashed 

1495: lines show the interval of averaging for the minimal response value

1496: $R_{min}$ in Fig.\ref{fig:CorrH}. Parameters of the chain

1497: are $s/r=34/21$, $K=5$ and $\tau _{0}=80$. }

1498: \end{figure}

1499: 

1500: 

1501: \begin{figure}[ht]

1502: \includegraphics[  width=115mm,

1503:   height=85mm,

1504:   keepaspectratio,

1505:   angle=90,

1506:   origin=c]{fig17.ps}

1507: \caption{\label{fig:CorrH} Dependence of the minimal response $R_{min}$ on $\hbar$

1508: for the parameters of Fig.\ref{fig:slide}. 

1509: }

1510: \end{figure}

1511: 

1512: In order to get a more quantitative picture, we estimate the value of the response

1513: function $R(i)$

1514: at its minimum in the middle of the chain by taking its average value inside the central region at $i=12\div22$: 

1515: $R_{min}=\langle R(i) \rangle_{i}$ (this interval is shown in Fig.\ref{fig:slide} by

1516: horizontal dashed lines). The dependence of $R_{min}$ on the parameter $\hbar$ 

1517: is presented in Fig.\ref{fig:CorrH}. We see, that the correlator in the central region 

1518: of the chain deviates from zero at $\hbar>2$. 

1519: According to the numerical data the response $R_{min}$ is very small for 

1520: $\hbar<\hbar_c\approx 2$ while it becomes rather strong for $\hbar>\hbar_c$.

1521: This confirms the existence of the quantum  phase transition 

1522: from the pinned to sliding phase at $\hbar=\hbar_c\approx 2$.

1523: \begin{figure}[ht]

1524: \includegraphics[  width=115mm,

1525:   height=85mm,

1526:   keepaspectratio,

1527:   angle=90,

1528:   origin=c]{fig18.ps}

1529: \caption{\label{fig:DEvsH} Dependence of the total

1530:  chain energy per particle 

1531: on $\hbar$ for the chain parameters $K=5$, $s/r=89/55$ and $\tau_0=80$;

1532: $\epsilon_0=5.302$ is an energy point chosen from convenience/illustration 

1533: reasons.}

1534: \end{figure}

1535: 

1536: What is the order of this transition? In our simulations we have repeated a cycle

1537: changing slowly $\hbar$ from $\hbar=1$ 

1538: to $\hbar=4$ and back about hundred times, but no hysteresis was found within

1539: the statistical errors. Indeed,   

1540: the difference of the chain energy per particle at the upward and backward paths 

1541: did not exceed 

1542: $10^{-3}$. This difference should be compared with the change of the energy per particle 

1543: $\delta E\approx 2.4$ which takes place when $\hbar$ changes from $\hbar=1$ to $\hbar=4$.

1544: The absence of hysteresis excludes  the phase transition of the first order. We 

1545: also do not see any breaks in the dependence of the chain energy on $\hbar$. 

1546: In order to make more visible small deviations 

1547: from a linear law we plot in Fig.\ref{fig:DEvsH} the quantity 

1548: $(\epsilon-\epsilon_0)/\hbar$, where $\epsilon_0=5.302$. The numerical data

1549: show  that the slope of energy dependence changes near $\hbar = \hbar_c \approx 2$.

1550: This change of slope is located  approximately at the same value of $\hbar$

1551: where the divergence of the correlation length takes place (see  Fig.\ref{fig:CorrX}).

1552: These data indicate that we have a second order quantum phase transition which appears

1553: near $\hbar_c\approx 2$.

1554: However, more extensive numerical simulations are required to determine more 

1555: precisely the order of the transition. 

1556: 

1557: 

1558: 

1559: \section{Conclusions}

1560: \label{sec:Conclusions}

1561: 

1562: We have studied quantum tunneling phenomena in a particular model

1563: of glassy material, the Frenkel-Kontorova chain. This system has

1564: a lot of states, which are exponentially degenerate and (meta)stable

1565: in the (quasi)classical limit. In the quantum case there are tunneling

1566: transitions between these states that can be understood in  terms 

1567: of instanton dynamics. In the quasiclassical limit at small $\hbar$ 

1568: the instanton density is small and instantons are local and isolated.

1569: With the increase of $\hbar $ their density grows and they start to overlap.

1570: 

1571: At sufficiently large $\hbar $ the instantons are coupled and

1572: become  collectivized. As a result, the tunneling of particles 

1573: in some fragments of the chain proceeds in correlated way.

1574: The size of these correlations grows until it reaches the size of the system.

1575: This leads to appearance of extended excitations

1576: and opening of a new gapless sliding phonon branch.

1577: Our data show that a quantum phase transition takes place

1578: between the pinned and sliding phases. Absence of hysteresis effects,

1579: as well as continuous dependence of chain energy on 

1580: $\hbar$ exclude the first order phase transition, so we can classify

1581: this transition as a continuous quantum phase transition. We stress

1582: that the quantum phase transition from pinned to sliding phase

1583: takes place in the regime where the classical chain always remains

1584: in the pinned phase with the finite phonon gap.

1585: 

1586: The direct analysis of Fourier spectrum of Feynman paths ensemble allowed

1587: to obtain detailed information on the dispersion law of low-lying 

1588: excitations in both quantum phases. Nevertheless, some questions remain open for further

1589: investigations. For example, one can analyze in more detail the effects 

1590: of interactions between instantons at low density and study their

1591: propagation properties in this regime. Another interesting remark 

1592: concerns the behavior of the system in the vicinity of the transition

1593: point at $\hbar_c\approx 2$. Indeed, in this region the kinetic

1594: energy per particle is approximately 0.6 that is about ten times smaller

1595: than the height of the potential barrier at $K=5$. Therefore more insights

1596: are required to understand the underlying physics of this transition.

1597:  

1598: One can ask on how general are our results obtained in the frame 

1599: of the Frenkel-Kontorova

1600: model? In fact, we never used any specific features of this model

1601: related to its nontrivial number theory properties. The only essential

1602: point is the existence  of an exponential number of quasidegenerate 

1603: states which is common for glassy materials and other disorder systems.

1604: Therefore it is very interesting to study an analogous quantum

1605: phase transition in systems with disorder and interactions.

1606: 

1607: This work was supported in part by the EC RTN network contract HPRN--CT-2000-0156;

1608: OVZ thanks the groups in Como and Toulouse for their hospitality

1609: during the work on this problem.

1610: %\begin{acknowledgments}

1611: %\end{acknowledgments}

1612: 

1613: %\bibliographystyle{revtex}

1614: %%%\bibliography{fk}

1615: 

1616: \begin{thebibliography}{99}

1617: \bibitem{FK38}Y.I. Frenkel and T.K. Kontorova, Zh. Eksp. Teor. Fiz. \textbf{8}, 1340 (1938)

1618: [Phys. Z. Sowjetunion \textbf{13}, 1 (1938)]. 

1619: \bibitem{Naba67} F. Nabarro, {\it Theory of Crystal Dislocations} (Clarendon, Oxford, 1967). 

1620: \bibitem{VPaAT84}V.L. Pokrovsky and A.L. Talapov, {\it Theory of Incommensurate Crystals}, 

1621: Soviet Scientifical Reviews Supplement Series Physics, vol.1 (Harwood, London, 1984). 

1622: \bibitem{SY71} S.C. Ying, Phys. Rev. B \textbf{3}, 4160 (1971). 

1623: \bibitem{Piet81}L. Pietronero, W.R. Schneider and S. Str\"{ a}sler, Phys. Rev. B \textbf{24},

1624: 2187 (1981). 

1625: \bibitem{Au78}S. Aubry, in \textit{Solitons and Condensed Matter Physics}, Eds. A.R.Bishop 

1626: and T.Schneider, (Springer, N.Y., 1978). 

1627: \bibitem{Au83a} S. Aubry, J. Phys. (France), \textbf{44}, 147 (1983).

1628: \bibitem{Flor96}L.M. Floria and J.J. Mazo, Adv. Phys. \textbf{45}, 505 (1996). 

1629: \bibitem{Brau97} O.M. Braun, T. Dauxois, M.V. Paliy, and M. Peyrard, Phys. Rev. Lett. 

1630: \textbf{78}, 1295 (1997); Phys. Rev. E \textbf{55}, 3598 (1997). 

1631: \bibitem{Cons00} L. Consoli, H.J.F. Knops, and A. Fasolino, Phys. Rev. Lett.

1632: {\bf 85}, 302 (2000).

1633: \bibitem{Borg89} F. Borgonovi, I. Guarneri and D. Shepelyansky, Phys. Rev. Lett.

1634: \textbf{63}, 2010 (1989).        

1635: \bibitem{Borg90} F. Borgonovi, I. Guarneri and D. Shepelyansky, Z. Phys. B \textbf{79}, 133 (1990).

1636: \bibitem{Borg89d} F. Borgonovi, Ph.D. thesis, Universita degli Studi di Pavia,

1637: Italia (1989).

1638: \bibitem{Berm94} G.P. Berman, E.N. Bulgakov and D.K. Campbell, Phys. Rev. B 

1639: \textbf{49}, 8212 (1994).

1640: \bibitem{Berm97} G.P. Berman, E.N. Bulgakov and D.K. Campbell, Physica D 

1641: \textbf{107}, 161 (1997).

1642: \bibitem{Hu98} B. Hu, B. Li and W.M. Zhang, Phys. Rev. E \textbf{58}, R4068 (1998).

1643: \bibitem{Hu99} B. Hu and B. Li, Europhys. Lett. \textbf{46}, 655 (1999).

1644: \bibitem{Hu01} B. Hu, B. Li and H. Zhao, Europhys. Lett. \textbf{53}, 342 (2001).

1645: \bibitem{Hu01r} B. Hu and  B. Li, Physica A \textbf{288}, 81 (2000).

1646: \bibitem{Ho01} C.-L. Ho and C.-I Chou, Phys. Rev. E \textbf{63}, 016203 (2001).

1647: \bibitem{Chir79} B.V.Chirikov, Phys. Rep. \textbf{52}, 263 (1979).

1648: \bibitem{Perc79} I.C.Percival, in {\it Nonlinear Dynamics and the Beam-Beam Interaction},

1649:         Eds. M.Month and J.C.Herra, AIP Conf. Proc., AIP, N.Y. {\bf 57}, 302 (1979).

1650: \bibitem{Au83b}S. Aubry, Physica D \textbf{7}, 240 (1983).

1651: \bibitem{Au83c}S. Aubry and P.Y. Le Daeron, Physica D \textbf{8}, 381 (1983).

1652: \bibitem{Mac84} R.S.~MacKay, J.~D.~Meiss and I.~C.~Percival, Physica D \textbf{13}, 55 (1984).

1653: \bibitem{Sch84} R. Schilling, Phys. Rev. Lett. {\bf 53}, 2258 (1984).

1654: \bibitem{Sch85} P. Reichert and R. Schilling, Phys. Rev. B \textbf{32}, 5731 (1985).

1655: \bibitem{Zhir01a} O.V. Zhirov, G. Casati and D.L. Shepelyansky, Phys. Rev. E \textbf{65}, 026220 (2002).

1656: \bibitem{Shur98} T. Shaefer and E.V. Shuryak, Rev. Mod. Phys. \textbf{70}, 323 (1998).

1657: \bibitem{Be80} H.U. Beyeler, L. Pietronero and S. Str\"{a}ssler, Phys. Rev. B \textbf{22}, 2988 (1980).

1658: \bibitem{PhonFK} O.V. Zhirov, G. Casati and D.L. Shepelyansky, 

1659: {\it Phonon localization in the Frenkel-Kontorova chain}, in preparation (2002).

1660: \bibitem{Poly77} A. Polyakov, Nucl. Phys. B \textbf{120}, 429 (1977).

1661: \bibitem{Creu81} M. Creutz and B. Freedman. Ann. Phys. \textbf{132}, 427 (1981).

1662: \bibitem{Shur84} E.V. Shuryak and O.V. Zhirov, Nucl. Phys. B \textbf{242}, 393 (1984).

1663: \bibitem{cQFT97} S.L. Sondhi, S.M. Girvin, J.P. Carini and D. Shahar, Rev. Mod. Phys. \textbf{69},

1664: 315 (1997).

1665: \bibitem{Metr53} N. Metropolis, A. Rosenbluth, M. Rosenbluth, A. Teller and E. Teller,

1666: J. Chem. Phys. \textbf{21}, 1087 (1953).

1667: 

1668: \end{thebibliography}

1669: 

1670: \end{document}

1671: